AI for Business: What It Does and Where to Start

Cover card with the title AI for Business: What It Does and Where to Start, in white text on a dark background with a cyan glow

AI for business means using software that can read, write and sort information to take over repetitive work. Start with one task that repeats every week, give an AI assistant your real context, check what it produces, and only then let it run on its own.

If you want to talk your own case through, the AI consulting page shows how I work with teams. Below: what AI does in a company, the three levels I use to think about it, what the data says and a 30-day start.

What does AI for business actually mean?

It means a tool that does one defined piece of work, such as drafting, summarizing, sorting or answering, while a person owns the result. Coursera defines it as “artificial intelligence solutions that specialize in making business processes more efficient” (page updated June 25, 2026), and most of the top guides say about the same thing.

Under the hood, most of it is machine learning: software trained on large amounts of data that learns patterns, then writes, sorts or predicts. You will meet it as chatbots for customers, drafting help in marketing, sorting and forecasting in operations, and assistants built into industry software. You don’t need to understand the math to use it, only to know which task it is doing.

Two cautions go with the definition.

First, the label gets stretched. A lot of vendors are just sticking the buzzword AI onto everything, and a lot of it isn’t really AI. Ask any vendor which exact task the product does and what a person still has to check.

Second, not every problem is an AI project. Sometimes it’s just an inventory system, and even a great system is pointless if people won’t use it. If a plain spreadsheet fixes the problem, use the spreadsheet.

How many businesses use AI today?

Fewer than the headlines suggest, and the number depends on who is counted. Two sources, two different answers:

  • U.S. Census Bureau, May 26, 2026. In its Business Trends and Outlook Survey, 19.8% of U.S. businesses used AI as of May 3, 2026. Use was 37% in firms with 250 or more employees and 32% in firms with 100 to 249. Under 20% of firms with four or fewer employees used it. By sector: about 40% in information, 33.9% in finance and insurance and about 14% in retail.
  • U.S. Chamber of Commerce, August 18, 2025. Its Empowering Small Business release says 58% of small businesses use generative AI, up from 40% in 2024.

The gap is big. A likely reason (my inference, not stated by either source): the Census asks whether AI is used in a business function, while the Chamber asks owners which technologies they use at work. One person asking a chatbot to draft an email may count in the second and not in the first. The Chamber is also an industry group, so read its release with that in mind.

What both agree on: use is rising, and most companies are still early. Starting now doesn’t make you a pioneer, and it doesn’t make you late. For the small-company version of this story, read AI for small business.

What are the three levels of using AI in a business?

Most companies sit on one of three levels, and each one needs a different amount of trust and checking. This is my own way of sorting it, not a standard from a study.

Level What it is Example Who checks Main risk
1. Ask You chat with an assistant, one task at a time Draft a reply, summarize a document, brainstorm names You, every time Average answers, because it knows nothing about your business
2. Give context The assistant gets your documents, rules and examples, and keeps them A saved set of instructions for how your company writes quotes You, on a sample Out-of-date or wrong context
3. Delegate The assistant does the task on a schedule or trigger and hands back finished work Sort every receipt overnight, write a morning brief from your inbox You approve anything that is written, sent or deleted A mistake repeated many times before anyone sees it

Level 1 is where everyone starts, and it has a ceiling. If you just ask an AI to build a website, make a logo or write social posts, the result will be average, because it takes the average of what it finds online. Mine gets better when I spend time giving it my knowledge and my taste. That’s Level 2, and it’s the biggest jump for most teams. I’ve spent over 1,500 hours mastering Claude (that’s my own count, not a study), and the main lesson is simple: stop repeating your rules in every chat and save them once. A fresh Claude chat knows nothing about you, your company or your rules, and my free Claude setup template fills that in for it.

Video: Setup Claude Better Than 99% Of People (Full Setup File) Video · 21 min Setup Claude Better Than 99% Of People (Full Setup File) I walk through my free setup file, which gives Claude your folders, memory and rules so every new chat starts knowing your business. Watch the video →

Level 3 is where the time comes back, and where the care goes. Here is what I do: I always give the assistant permission to read, because that does no harm, but I never give it blanket permission to edit or delete. When it wants to write, edit or delete, it asks and I approve. There are horror stories of an AI that deleted someone’s files.

When you teach it a process, watch it the first few times. If it goes the wrong way you say stop, fix it, and keep going. After two, three or four runs you can tell whether it still needs you.

Don’t skip levels. A company that jumps straight to Level 3 with no Level 2 gets fast, confident, average output. If you want the longer version of Level 3, see AI agents for business and AI automation, or have me build it through my custom AI apps service.

Which jobs does AI do well, and which should it skip?

It does well on text-heavy work that repeats and is cheap to check. If someone does the same task over and over every day, like sending receipts or emails, that task can probably be automated. It does poorly where a wrong answer is costly and hard to spot.

Kind of work Good fit? Why A first test
Drafting emails, quotes and replies Yes Easy to read before it goes out Draft ten routine replies and compare time per reply
Summarizing calls, documents and notes Yes Quick to check against the source Summarize five calls and ask attendees what it missed
Sorting receipts, forms and requests Yes, with a sample check Repetitive, and mistakes show up in totals Run last month beside your usual process
Research and comparing options Yes, with care It’s fast but can state wrong facts confidently Ask several tools the same question and compare
Legal wording, tax filing, medical advice No, not alone A wrong answer is costly and hard to spot Keep a qualified person in charge
Anything customers see with no review Not yet The assistant speaks for you Review a few hundred answers before you trust it

On the research row, when I need a list of prospects or investors, I ask ChatGPT, Gemini, Claude, Perplexity and the other AI models the same prompt and compare the answers. Comparing the answers shows you quickly where a single tool might be off.

For a department-by-department menu, read AI use cases for business. For the tools themselves, see AI tools for small business and AI software for business.

What do the results look like in real research?

One of the best-known studies is on customer support. In a paper by Erik Brynjolfsson, Danielle Li and Lindsey Raymond, NBER working paper 31161, a generative AI assistant was rolled out to 5,179 customer support agents. Productivity, measured as issues resolved per hour, rose about 14% on average. For novice and low-skilled workers it rose about 34%, while experienced workers saw minimal impact.

That is one company’s setting, so read it as a reason to try an assistant on support work, not as a forecast for your business. It also hints at where the value sits: the tool helped least-experienced people the most, so the training and examples you give it matter.

What can go wrong?

Four things, and each is cheap to handle before you start.

Confident wrong answers. How many times have you told an AI “no, that’s wrong” and heard “You’re so right, I’m really sorry”? Eventually it will get to where it’s rarely wrong, but we aren’t there yet. Keep a person in the loop for anything a customer, bank or regulator will see.

Your data. Decide what must never go into an AI tool, such as payment details, health information, passwords and contracts. Put it in a one-page rule. An AI policy that only limits people makes them afraid to do anything with AI, so say what is allowed as clearly as what isn’t. The post on AI governance has the longer version.

Hype. On September 25, 2024, the U.S. Federal Trade Commission announced Operation AI Comply, five enforcement actions against deceptive AI claims. Its chair said there is “no AI exemption from the laws on the books.” Be wary of anyone who promises income or savings from AI with no method behind it.

Over-building. Your first attempt will probably be bad, and that’s normal. I once spent twelve hours on a Gmail skill, which is a saved set of instructions for the assistant, and had to scrap the whole thing. I learned a lot from it. The fix is to start with a small task, get it working by hand and only then automate it.

Do you need a technical person?

Not to start. You don’t have to be a coder or a developer. I’m not one, and I still use these tools. What you need is someone who knows the work well enough to say whether the output is right.

Technical help becomes useful when a task has to connect to your accounting, customer or scheduling systems. Before then, one owner, one task and a few hours a week are enough. Plan some practice for the team too: see AI training for employees, AI prompts and my live AI workshops.

How do you start in 30 days?

Follow five steps, in order. If you skip the first, you have nothing to measure the rest against.

  1. Pick one task and one number. Write it in a sentence: “Cut the time we spend writing quote follow-ups from five hours a week to two.” My goal is really small, easy, quick wins. When a team sees how fast something gets built, they want to automate everything else.
  2. Pick one tool. Check the software you already pay for first. If it has an assistant that covers the task, test that. If not, choose one general assistant and the cheapest plan that lets you test.
  3. Record a baseline. Write down the hours, the error rate or the replies per day before you start.
  4. Run it for four weeks at Level 1 or 2. One person owns it. Save the instructions you give it so a colleague can repeat them. Set a stop rule in advance: “If we don’t save at least X hours a week, we stop.” Stopping is a result, not a failure.
  5. Decide with the numbers. Widen to a second task, change the setup or stop. Only then think about Level 3. AI strategy helps you rank the next three tasks, and AI ROI shows how to count the return.

The four-week length is a practical suggestion, not a research finding.

When should you hire help?

Hire help when the work moves past a pilot: the task has to connect to other systems, it touches customer data, or a pilot stalled and nobody has time to find out why. On a first call, you tell me about your business and your goals. I ask questions and tell you honestly whether I can help, and what I would do first.

Before you hire anyone, read what an AI consultant does and run the AI readiness assessment so you know where you stand. When you’re ready to talk, start on the consulting page.

Updated November 18, 2026. Sources opened on that date: the U.S. Census Bureau story of May 26, 2026, the U.S. Chamber of Commerce release of August 18, 2025, NBER working paper 31161, the FTC announcement of September 25, 2024 and Coursera’s AI for business page. The three levels are my own framework, not a published standard.

FAQs

What does AI for business mean?

It means using software that can read, write, sort and analyze to take over repetitive work in a company, such as drafting replies, summarizing calls or sorting receipts. A person still owns the result.

What is the 30% rule for AI?

There is no official rule. The number comes from a loose idea that AI handles about 70% of repetitive work and people handle the rest. Coursera repeats it without naming a source, so treat it as a rule of thumb, not a standard, and measure your own tasks.

Can I make $1,000 a day using AI?

Nobody can promise that, and promises like it are a warning sign. In September 2024 the U.S. Federal Trade Commission announced five actions against deceptive AI claims, including income promises. Judge any course or tool by the hours it saves you on work you already do.

Do I need to know how to code to use AI in my business?

No. Asking a chat assistant for help, giving it your documents and checking what it writes needs no code. Code skills matter only later, when a task has to connect several systems.

Which AI should a business start with?

One general assistant such as ChatGPT, Claude or Gemini, or the assistant already inside software you pay for. Pick one, test it on one task for a month and judge it by your own before-and-after numbers.

Get new resources in your inbox

Consulting

Grow your business with a founder who’s done it

David coaches entrepreneurs and advises companies as a consultant and fractional Chief AI Officer.

Speaking

A speaker your audience will remember

Motivational keynotes & workshops on entrepreneurship, AI, financial literacy, and more.